Skip to content

FRM Exam Part II · Structured Credit Risk

Default Correlation and the Gaussian Copula Explained

Updated 11 October 2026 · Fact-checked

Default correlation is the tendency of borrowers to default together. The one-factor Gaussian copula links each firm's latent variable to a common factor and an idiosyncratic term, then sets default when the variable falls below a threshold. Higher correlation raises equity tranche value and lowers senior tranche value.

Understand Default Correlation and the Gaussian Copula

Default correlation measures how likely borrowers are to default at the same time. If defaults were independent, a large pool would lose close to its expected loss every year. In reality, a weak economy hits many borrowers together, so losses cluster. This clustering is what makes tranche risk different from pool risk.

The one-factor Gaussian copula is the standard way to model this. Each firm i has a latent variable: Xi = √ρ × M + √(1 − ρ) × Zi. M is a common (market) factor. Zi is a firm-specific factor. Both are independent standard normals. The firm defaults by time T if Xi falls below a threshold K = N⁻¹(PD). Here PD is the default probability to time T and N⁻¹ is the inverse standard normal CDF.

The copula joins the individual default probabilities into a joint distribution. The marginal PDs are fixed by the thresholds. The correlation ρ only controls how defaults bunch together. Conditional on M, defaults are independent. This makes the model easy to compute for large pools.

The conditional default probability given M is: PD(M) = N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)]. A bad market draw (low M) lifts this probability for every firm at once.

Correlation shifts risk between tranches without changing the pool's expected loss. Higher correlation makes both very few and very many defaults more likely. The equity tranche takes the first losses, so it gains from the higher chance of zero or few defaults. Its value rises with correlation. The senior tranche only loses in a severe, widespread default event. Higher correlation makes that more likely, so its value falls. The mezzanine tranche can go either way and is often not monotonic.

Market quotes use two conventions. Compound correlation is the single correlation that reprices one tranche. It can be non-unique for mezzanine tranches, and may have no solution. Base correlation treats each tranche as the difference between two equity tranches with the same lower attachment of zero. Each equity tranche has a unique correlation. Base correlation therefore avoids the non-uniqueness issue and is more widely used. Quoted base correlations usually differ by detachment point, a pattern called the correlation skew, which shows the flat-correlation Gaussian copula does not fit the market.

Key formulas to remember

One-factor Gaussian copula
Xi = √ρ × M + √(1 − ρ) × Zi
M and Zi are independent standard normals. Pairwise asset correlation is ρ.
Default threshold
K = N⁻¹(PD)
Firm defaults if Xi < K. Marginal PD is unchanged by ρ.
Conditional default probability
PD(M) = N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)]
Defaults are independent given M. Low M raises PD(M).
Joint default probability of two firms
P(both default) = PD² + ρD × PD × (1 − PD), if both have the same PD
ρD is the default correlation (default indicators). It is usually smaller than the asset correlation ρ.
Tranche loss rule
Tranche loss = min(max(L − A, 0), D − A)
L is pool loss, A attachment, D detachment, all in the same units.
Correlation effects
Equity value ↑ and senior value ↓ as ρ ↑
Pool expected loss is unchanged. Mezzanine is ambiguous.
Base correlation tranche
Tranche A–D = equity(0–D) − equity(0–A)
Each equity piece uses its own base correlation.

How to solve Default Correlation and the Gaussian Copula questions

Use this method for any question on default correlation, copula mechanics or tranche value.

  1. 1Identify what is asked: a copula calculation, a direction of change in tranche value, or a correlation convention.
  2. 2Fix the marginal PDs first. The copula never changes them. Compute K = N⁻¹(PD) if a threshold is needed.
  3. 3If a conditional PD is asked, substitute ρ and the given M into the PD(M) formula. Compute the numerator, then divide by √(1 − ρ).
  4. 4Convert the result using the standard normal table. Check that a lower M gives a higher PD(M).
  5. 5For tranche questions, note the attachment and detachment points. Ask whether losses hit the tranche in a few defaults or only in many defaults.
  6. 6Apply the rule: equity gains from higher ρ, senior loses, mezzanine depends on its position. Expected pool loss does not change.
  7. 7For correlation convention questions, state whether the tranche is priced alone (compound) or as a difference of equity tranches (base).
  8. 8Check the answer for sense: probabilities between 0 and 1, and loss never above the tranche width.

Quickest way: Direction-of-effect shortcut

When to use it: Use when the question asks how a rise or fall in correlation changes a tranche value or spread.

  1. Ask: does this tranche lose in the first few defaults (equity) or only in a mass-default event (senior)?
  2. Higher correlation: equity value up, equity spread down. Senior value down, senior spread up.
  3. Lower correlation: reverse both.
  4. For mezzanine, say the effect is ambiguous unless the question gives details.
  5. For base vs compound, pick base if the question stresses uniqueness or equity-tranche building blocks.

Common mistakes in Default Correlation and the Gaussian Copula

  • Saying higher correlation raises the expected loss of the pool.

    Students link correlation to riskier, so they assume more loss.

    Fix: Expected pool loss depends only on PD, LGD and exposure. Correlation changes the shape of the loss distribution, not its mean.

  • Claiming senior tranches gain from higher correlation.

    Confusing diversification with tranche position.

    Fix: Higher correlation fattens the tail. Senior tranches only lose in the tail, so they lose value.

  • Treating asset correlation ρ as the same as default correlation.

    Both are called correlation in the copula setup.

    Fix: ρ is the correlation of latent variables. Default correlation is the correlation of default indicators, usually lower.

  • Forgetting the √(1 − ρ) divisor in the conditional PD formula.

    Students memorize only the numerator.

    Fix: Write the whole formula first: N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)]. Check the sign of M.

  • Saying compound correlation is always unique.

    Assuming one tranche price gives one correlation.

    Fix: Mezzanine value is not monotonic in correlation, so compound correlation may have two solutions or none. Base correlation avoids this.

Worked examples

Example 1

In a one-factor Gaussian copula, a firm has a 5-year PD of 2.28% so that N⁻¹(PD) = −2.00. Asset correlation ρ = 0.36 (√ρ = 0.6, √(1 − ρ) = 0.8). Find the conditional default probability if the market factor M = −2. Use N(−0.50) = 0.3085, N(−1.00) = 0.1587, N(0.00) = 0.5000, N(0.50) = 0.6915.

Show the solution
  1. Use PD(M) = N[(N⁻¹(PD) − √ρ × M) ÷ √(1 − ρ)].
  2. Compute √ρ × M = 0.6 × (−2) = −1.2.
  3. Numerator = −2.00 − (−1.2) = −0.80.
  4. Divide by 0.8: −0.80 ÷ 0.8 = −1.00.
  5. PD(M) = N(−1.00) = 0.1587.

Answer: The conditional default probability is about 15.87%, far above the unconditional 2.28%, because the market factor is bad.

Example 2

A CDO has an equity tranche (0–3%), a mezzanine tranche (3–10%) and a senior tranche (10–100%). The market correlation assumption rises from 0.20 to 0.40, with all PDs unchanged. Which statement is correct?
A. Equity value falls and senior value rises.
B. Equity value rises and senior value falls.
C. Both equity and senior values rise.
D. Both fall because pool expected loss rises.

Show the solution
  1. Pool expected loss depends on PDs, which are unchanged. So D is wrong.
  2. Higher correlation makes zero or few defaults more likely, which helps the equity tranche that takes the first losses.
  3. Higher correlation also makes mass defaults more likely, which hurts the senior tranche that is hit only in that tail.
  4. So equity value rises and senior value falls. Option B.

Answer: B. Equity value rises and senior value falls.

Exam tips

  • Memorize the direction: correlation up means equity value up, senior value down, pool expected loss unchanged.
  • In a calculation, write the conditional PD formula before plugging numbers. Check the sign on M.
  • Watch the wording: asset correlation, default correlation and base correlation are different things.
  • If a question mentions correlation skew, link it to the failure of a single flat correlation to fit all tranches.
  • On a phone-style quick read, find the tranche position first, then answer.

Practice questions from Structured Credit Risk

Default Correlation and the Gaussian Copula in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Default Correlation and the Gaussian Copula: frequently asked questions

What does the Gaussian copula do in credit modelling?

It joins individual default probabilities into a joint default distribution using a correlation parameter. The marginal PDs stay fixed. The correlation only controls how often defaults occur together.

How does correlation affect equity and senior tranche value?

Higher correlation raises equity tranche value because zero or few defaults become more likely. It lowers senior tranche value because a severe, widespread default event becomes more likely. Mezzanine tranches can move either way.

What is the difference between base correlation and compound correlation?

Compound correlation is the single correlation that reprices one tranche on its own. Base correlation prices each tranche as a difference of equity tranches starting at zero. Base correlation is unique for each equity piece, while compound correlation may not be.

Is default correlation the same as asset correlation in the copula?

No. Asset correlation ρ links the latent variables. Default correlation is the correlation of the default events, and it is generally lower than ρ.